Best ImageMagick parameters for rescaling images with text

[Doitsu]

One of my ebooks contains many .png images of Arabic text that was saved with a relatively small font size (most likely 11 or 12pt), which makes it really hard to read. (For an example, see the attached file.)

I wondering if there are any ImageMagick* parameters that do the following:

1. Enlarge the picture by about 25-50%.
2. Smooth the resulting image.

Basically, I'm looking for any parameters that'll make the image slightly larger and more readable.

*It doesn't have to be ImageMagick; any suitable tool that supports batch processing is fine.

[BetterRed]

@Doitsu - if you have access to Windows have a look at ImBatch ==>> http://www.highmotionsoftware.com/products/imbatch

I have about half a dozen multi-step imbatch jobs that I use regularly on sets of many hundreds of images, they save me heaps of time.

BR

[Tex2002ans]

Is this just going to be for your own personal usage? Or is this for an actual book?

Was this a bunch of inline images instead of Arabic text? (I know I have run across similar problems with crappy conversions of Greek text.)

As you know, trying to UPSCALE things will result in some subpar images... but if you just needed something a little quick/dirty, I dabbled in this for a little and came up with this basic workflow.

Step 1

Original Image:



First you upscale the image:

convert test2.png -scale 1000% test2[Scaled1000].png

Note: You can insert any % there. 1000% makes it 10x larger.

Step 2

Then you can adjust the contrast slightly:

convert test2[Scaled1000].png -contrast-stretch 3% test2[ContrastTweak3].png

Note: You can insert any % there. I found that anywhere from 1-5% seemed decent. Going beyond that distorted the text too much.

Step 3

Then you can scale the image down:

convert test2[ContrastTweak3].png -scale 40% test2[ScaledDown40].png

Note: You can use whatever % or size you want here. This results in a final image that is 4x larger.

Step 4

If you wanted to convert on an entire folder of PNGs, just go to the folder and run this:

convert *.png -scale 1000% -contrast-stretch 3% -scale 40% -set filename:fname %t_tn +adjoin %[filename:fname].png

Note: If you have a ton of images, probably better to run mogrify or come up with some sort of batch file. I read about the convert -set filename method here: https://www.imagemagick.org/Usage/ba...ogrify_convert

You could also probably introduce more steps in between (different resizing filters, etc. etc.), but I am not that familiar with that. Most of the time I am just using Imagemagick to clean up speckling in scans. :P

[Doitsu]

@Tex2002ans: Thanks for your detailed instructions!
@BetterRed: Thanks for the ImBatch recommendation! I checked it out and I especially liked the built-in preview mode.

[BetterRed]

@Doitsu - I especially like the ability to see the effect when tasks are re-ordered. Resize then sharpen ain't the same as sharpen then resize.

FYI the images I process are from fixed cameras on 'wildlife' trails. Objective, eradication of feral animals - pigs, horses, goats, rabbits, foxes, hares, mice, rats etc. That's why there's hundreds of images - one scampering rat can trigger a dozen shots or more.

BR

[willus]

Originally Posted by Tex2002ans

...

convert *.png -scale 1000% -contrast-stretch 3% -scale 40% -set filename:fname %t_tn +adjoin %[filename:fname].png

...

Thank you for this detailed answer. After reading about how the contrast-stretch option works in ImageMagick, I've learned something new. I may change the way k2pdfopt does contrast adjustment now. I'll have to try a few test cases.

[Tex2002ans]

Originally Posted by willus

Thank you for this detailed answer. After reading about how the contrast-stretch option works in ImageMagick, I've learned something new. I may change the way k2pdfopt does contrast adjustment now. I'll have to try a few test cases.

I will have to research this topic/methods a lot more as well.

That method was just something I came up with in a few hours as a ROUGH resizing (better than nothing, but definitely not something I would use in an ebook for sale). There has to be much better Imagemagick solutions that exist out there.

Most of the time when I run across Doitsu's problem in all the books I work on is microscopic images of Greek text... but instead of me doubling/tripling the size, I just turn it into the actual Unicode characters instead. :P

* * * * *

I remember when I was trying to come up with a despeckling solution to clean up PDF scans. I spent days reading everything on Google, I somehow stumbled upon a REALLY obscure Imagemagick post that had the (close-to-perfect) solution.

If you did a search for it in Google, it was one of those where you get "6 hits" and/or it was buried on "page 1000" of a normal search (like "Imagemagick despeckling" or "cleaning speckles in PDF scans").

If I remember correctly, they did it by shrinking, then expanding pixels, then creative a diff image. Then step through and remove any group of non-fully-black pixels that was smaller than X x Y. This would get rid of little 1x1, 1x2, 2x1, 2x2, 3x2, [...] specs of dust all over the page.

Then it would reverse the expansion and use the diff image to try to return it to its original form.

It did a pretty fantastic job compared to ANYTHING I ever found before/since... BUT the downfall was sometimes the fine details would disappear... like the very tip of a comma, or the tip of a serif font, or little umlaut above a character in a footnote, or a badly scanned period would just poof.

Or complicated edges/lines would go poof (like a dashed line in a graph, or a very thin line separating the bottom of the text from the footnote section).

Definitely helped me in many cases though, and helped cut down on the size of some speckled PDFs/Images DRAMATICALLY.

And you had to weigh what would be worse... a whole PDF full of little speckles? Or the tiny end of a comma or an accent disappearing. :P

* * * * *

Then you get the REALLY obscure/ingenious stuff... like using ImageMagick to add line numbers to a PDF:

https://tex.stackexchange.com/questi...mbers-to-a-pdf

I mean look at that top answer... genius.

Coming up with super genius ways to compare the average color of a line, to tell if it was actual text or "blank space".

[dgatwood]

Given that these are presumably generated with a standard Arabic font, it might be possible to do some really smart OCR and turn it into actual text.

[Doitsu]

Originally Posted by dgatwood

Given that these are presumably generated with a standard Arabic font [...]

They were generated with a standard Arabic font, however, the font contains several (optional) ligatures that mimic traditional Arabic handwriting, which makes Arabic OCR of these particular images rather difficult.

For an example, have a look at the attached screenshot. It shows the same word in two different fonts (with and without ligatures).

To give you an idea of the difficulties, here's the same word split into the 5 letters that the OCR program would have to detect:

ﺟ ﺮ ـﻳ ـﻣ ﺔ